Elamur Research⌕
☰
home / papers / 2026 / A hybrid approach to predicting the age group of authors of texts written in Portuguese

A hybrid approach to predicting the age group of authors of texts written in Portuguese

Original (Portuguese): Uma Abordagem Híbrida para Predição de Faixa Etária de Autores de Textos Escritos na Língua Portuguesa

Paper · 2026 · Alice Rezende Ribeiro and Luiz Henrique de Campos Merschmann

Summary

Inferring an anonymous author's age group from Portuguese text is still hard. Multiclass age prediction in prior work reaches only about 0.41–0.67 F1, while gender prediction exceeds 0.9, and Portuguese has far fewer resources than English. Ribeiro and Merschmann propose a two-stage hybrid pipeline. Texts are first cleaned by removing stopwords, punctuation and domain-specific tokens such as HTML tags or hashtags, then lowercased. A dictionary-based module then tries to assign an age group. It uses weighted word lists derived from the coefficients of a linear SVM trained on TF-IDF features. This extends the single binary gender lexicon of Sap et al. (2014) to multiple dictionaries for a multiclass problem. When the dictionaries cannot decide, a conventional multiclass classifier makes the prediction. The authors compared this against the strongest published results for each evaluated dataset. The hybrid approach always matched or outperformed them. This supports the hypothesis that domain-specific vocabulary helps age profiling. The method is language-independent, which makes it potentially transferable to other low-resource settings.

Key points

  • Baselines to beat: On Portuguese data, earlier multiclass age prediction scored 0.62 F1 on B5 Corpus, 0.45 on BlogSet BR and 0.41 on BR Moral (Dias, 2019). Flores et al. (2022) reached 0.67 weighted F1 with an LSTM on e-SIC citizen requests.
  • Dictionary module: The lexicons come from a linear SVM trained on TF-IDF values instead of relative word frequencies. There is one weighted dictionary per age class, and each word's coefficient indicates how strongly it is associated with that class.
  • Cascade logic: The traditional classifier is used only as a fallback, for texts where the dictionary stage gives no decision.
  • Outcome: Across all evaluated datasets, results were better than or equal to the best prior approaches. The exact scores fall in the omitted part of the extracted text.
  • Future work: The authors plan to test the method on other languages and to find dictionary construction strategies that increase the coverage of the first stage. The cited tooling includes scikit-learn, Keras and NLTK.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Alice Rezende Ribeiro and Luiz Henrique de Campos Merschmann. 2026. Uma Abordagem Híbrida para Predição de Faixa Etária de Autores de Textos Escritos na Língua Portuguesa. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{ribeiro2026uma,
  title = {{Uma Abordagem Híbrida para Predição de Faixa Etária de Autores de Textos Escritos na Língua Portuguesa}},
  author = {Alice Rezende Ribeiro and Luiz Henrique {de Campos Merschmann}},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.31/}
}

References

20 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Janek Bevendorff, Berta Chulvi, Elisabetta Fersini, Annina Heini, Mike Kestemont, Krzysztof Kredens, Maximilian Mayerl, Reynier Bueno, Piotr Pezik, Martin Potthast, Francisco Rangel Pardo, Paolo Rosso, Efstathios Stamatatos, Benno Stein, Matti Wiegmann, Magdalena Wolska, e Eva Zangerle. 2022. Overvi… find
  2. Steven Bird, Ewan Klein, e Edward Loper. 2009. Natural Language Processing with Python. O’Reilly Media, Inc., Sebastopol, CA. find
  3. François Chollet. 2015. Keras. https://keras.io. ↗
  4. José Pereira Delmondes Neto. 2021. Caracterização autoral interdomínio a partir de textos. Dissertação, Escola de Artes, Ciências e Humanidades, Universidade de São Paulo, São Paulo. find
  5. Rafael Felipe Sandroni Dias. 2019. Caracterização autoral a partir de textos utilizando redes neurais artificiais. Dissertação de mestrado, Escola de Artes, Ciências e Humanidades, Universidade de São Paulo. find
  6. Fabio Duarte. 2025. Amount of data created daily. https://explodingtopics.com/blog/ data-generated-per-day. Acessado em: 09 de agosto de 2025. ↗
  7. Arthur Marçal Flores, Matheus Camasmie Pavan, e Ivandré Paraboni. 2022. User profiling and satisfaction inference in public information access services. Journal of Intelligent Information Systems, 58(1):67–89. find
  8. Rita Georgina Guimarães, Renata Lopes Rosa, Denise De Gaetano, Demóstenes Zegarra Rodríguez, e Graça Bressan. 2017. Age groups classification in social network using deep learning. IEEE Access, 5:10805– 10816. find
  9. Fernando Hsieh, Rafael Dias, e Ivandré Paraboni. 2018. Author profiling from facebook corpora. Em Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find
  10. João Morais e Luiz Merschmann. 2022. A cascade approach for gender prediction from texts in portuguese language. Em Proceedings of the 28th Brazilian Symposium on Multimedia and the Web, páginas 151–158, Porto Alegre, RS, Brasil. SBC. find
  11. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, e E. Duchesnay. 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12:2825–… find
  12. Balla Charishma Sulochana, Bhavya Sri Pragada, Boya Chaitanya Kiran, Gaddam Anvith Reddy, e Manju Venugopalan. 2024. Author identity unveiled: Gender and age prediction from textual patterns using bert. Em 4th International Conference on Intelligent Technologies, páginas 1–6. find
  13. Szde Yu. 2023. Cyber profiling: Predicting political orientation with socmint. Telematics and Informatics Reports, 10:100058. find
  14. Ani Petrosyan. 2025. Internet usage worldwide -statistics & facts. https://www.statista.com/ topics/1145/internet-usage-worldwide/ \#topicOverview. Acessado em: 09 de agosto de 2025. ↗
  15. Ricelli Ramos, Georges Neto, Barbara Barbosa Claudino Silva, Danielle Sampaio Monteiro, Ivandré Paraboni, e Rafael Dias. 2018. Building a corpus for personality-dependent natural language understanding and generation. Em Proceedings of the 11th International Conference on Language Resources and Eval… find
  16. Anne-Sophie Riegger, Jan F. Klein, Katrin Merfeld, e Sven Henkel. 2021. Technology-enabled personalization in retail stores: Understanding drivers and barriers. Journal of Business Research, 123:140– 155. find
  17. Henrique dos Santos, Vinicius Woloszyn, e Renata Vieira. 2018. Blogset-br: A brazilian portuguese blog corpus. Em Proceedings of the 11th International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find
  18. Wesley Santos e Ivandré Paraboni. 2019. Moral stance recognition and polarity classification from twitter and elicited text. Em Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2019), Varna, Bulgaria. INCOMA Ltd. find
  19. Maarten Sap, Gregory Park, Johannes Eichstaedt, Margaret Kern, David Stillwell, Michal Kosinski, Lyle Ungar, e H. Schwartz. 2014. Developing age and gender predictive lexica over social media. Em Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), páginas…
  20. Douglas Henrique Silva. 2020. Classificação de gêneros e faixas etárias em redes sociais online por meio de técnicas de aprendizagem multidimensional. Dissertação de mestrado, Universidade Federal de Lavras. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ribeiro-2026-uma-abordagem-hibrida-para-predicao/